Prev. Article Next Article Table of Contents

Research Article

ART 2-A for Optimal Test Series Design in QSAR^§

Daniel Domine,*^† James Devillers,^† Dietrich Wienke,^‡ and Lutgarde Buydens^‡

CTIS, 21 rue de la Bannire, 69003 Lyon, France, and Catholic University of Nijmegen, Laboratory for Analytical Chemistry, Toernooiveld 1, 6525 ED Nijmegen, The Netherlands

J. Chem. Inf. Comput. Sci., 1997, 37 (1), pp 10–17

DOI: 10.1021/ci960376p

Publication Date (Web): January 27, 1997

Key words: adaptive resonance theory; ART networks; hierarchical cluster analysis; nonlinear mapping; nonlinear neural mapping; selection of test series.

Author to whom all correspondence should be addressed.

†

CTIS.

‡

Catholic University of Nijmegen.

Abstract

The family of adaptive resonance theory (ART) based systems concerns distinct artificial neural networks for unsupervised and supervised clustering analysis. Among them, the ART 2-A paradigm presents numerous strengths for data analysis. After a rapid presentation of the ART 2-A theory and algorithmic information, the usefulness of this neural network for the selection of optimal test series is estimated. The results are compared with those obtained from hierarchical cluster analysis and visual mapping methods. The advantages and drawbacks of each method are discussed. We show that ART 2-A represents a new useful nonlinear statistical tool for QSAR and drug design.

View: Full Text HTML | Hi-Res PDF